Parallel imaging uses the spatial sensitivity profiles of multiple receive array elements to recover images from undersampled k-space data, allowing faster acquisition without image quality penalty (at the cost of some SNR).

The Basic Principle

If k-space is undersampled by an acceleration factor R (only every Rth phase-encode line is acquired), the Nyquist criterion is violated and aliasing (fold-over artifacts) occurs in the image. However, if multiple receive channels are available, each with a different spatial sensitivity, the aliased signals from different locations can be separated using the known sensitivity patterns.

SENSE

Sensitivity Encoding (SENSE) operates in image space. The aliased image from each channel is combined using a matrix inversion based on the measured coil sensitivity maps. Requires explicit sensitivity calibration and produces a noise amplification characterised by the geometry factor (g-factor).

GRAPPA

GeneRalised Autocalibrating Partial Parallel Acquisition (GRAPPA) operates in k-space. Missing k-space lines are synthesised from acquired lines using calibration coefficients estimated from a central fully-sampled auto-calibration signal (ACS) region. More robust to motion between calibration and imaging than SENSE.

SNR and g-Factor

Parallel imaging reduces scan time by factor R but always reduces SNR by at least \(\sqrt{R}\) (less data = more noise). The g-factor (geometry factor) \(\geq 1\) further penalises SNR due to the conditioning of the unaliasing matrix. \(SNR_{PI} = SNR_{full} / (g\sqrt{R})\). High-field MRI (7 T) benefits most from parallel imaging because the SNR advantage at high field partially compensates for the \(\sqrt{R}\) SNR loss.